SerialSeb 🇲🇨🇪🇺🏳️‍🌈

some arseholes are trying to steal money from my family and friends. im at home safe and do not need or have I ever asked for money. call the police immediately if contacted.

@jlongster OpenAI compat providers switching to stateful sessions. I built a proxy in the meantime but the less code I build the better :)

@serialseb what do you mean?

OpenCode is about to get more powerful with remote sandboxes I showed a brief demo before, but here's a much more in-depth demo. it's not hard to add basic support for a remote env, but handling all the edge cases like when a remote env gets deleted is difficult. especially if care about good UX You never want to lose session data. so the choices are: run the session in your env, but run all tool calls remotely. that's too complex and painful. The other way is to just let the full session run remotely, but sync back all the session data in your env. We chose this path: we built a syncing system which logs all events in a way that we can always recreate your entire session. That means the remote env could get destroyed, but we can easily restore it. it also opens up other interesting ideas which we'll be exploring

Just let Sonnet work for half an hour on horrible code telling them “I’ll tell you later”. Funny to see it trying to figure out visitor patterns and dual dispatch for lock free data passing. Lolz. It’s trying poor thing.

@bertyJobbo @Aaronontheweb 5.4 is more like a Labrador, “code? Code? Where’s my code? Please give me the code, I want the code!” Two minutes later “maybe I should rewrite everything without committing anything first”

@Aaronontheweb 5.3 is such a weird model. Very powerful but almost like a grumpy teenager with the "oh you want _me_ to do it? Why didn't you say?"

Have ChatGPT / Codex subscriptions working with Netclaw. Codex-5.3 by default is an extremely lazy model. Would not call tools it had access to unless I explicitly instructed it to, constantly asked for permission, etc. Compare this to Qwen3.5 which just does it

@hhariri @jetbrains That’s the one. Such good times and good memories. Even if I was a lil sh*t back then :)

@serialseb @jetbrains The Golden Lion. How could one forget :).

You want a great company to do AI with, go for @jetbrains, in decades they hve been nothin but honest, empathic, and just a fab bunch of people (anyone remembers that pub in I think it was Malmö?) /cc @hhariri

@mitchellh @jezell As I’m writing a stupid TUI for fun, are there numbers of the max FPS / bps from stdout for Gostty? It’s my only franme of reference as its my only tool but wondered if there was documentation (I looked and didn’t find)

Happy to share that we've signed 5 contributor contracts for Ghostty totaling ~350 committed hours (~$21k) covering community management, graphics, Unicode compat, and GTK. This is a big milestone, Ghostty is paying contribs for the first time! ghostty.org/docs/sponsor

Thanks @AnthropicAI

And that’s my @claudeai subscription on the way for cancellation an payments on their way to be reverted on my credit card. Love the model, the company… Well.

Anthropic discovered that Claude Opus 4.6 was cheating during the BrowseComp benchmark. > On one question it spent ~40M tokens searching before realizing the question looked like a benchmark prompt. > The model then searched for the benchmark itself and identified BrowseComp. > It located the evaluation source code on GitHub, studied the decryption logic, found the encryption key, and recreated the decryption using SHA-256. > Claude then decrypted the answers for ~1200 questions to get the correct outputs. > This pattern appeared 18 times during evaluation. > Anthropic disclosed the issue publicly, reran the affected tests, and lowered their benchmark scores. Respect for the transparency 🫡🫡🫡

The lack of security in @AnthropicAI is not just worrying, it’s against some regulations on credit card payments these days I believe. We are on 2005? Please everyone use apple login protect yourself and your data.

I’m not ready to commit myself to anyone but, if I was, I think it would be @AnthropicAI Opus. #justsaying

@migueldeicaza I was in a debate on w=2 and exception tables for my new TUI with chatGPT :) just a bit of fun before I go open source. Perfect companion to Ghostty :)

@serialseb Wait what?

@migueldeicaza I think a certain model has a crush on you…

When you have terrible insomnia… it’s the moment to push codex 5.3 to its limits. 240k tokens so far. Holding well.

@mitchellh thank you for making alt screen look nice when we paint. Or it’s an accident. Either way it looks great. Also zero width what’s the deal? :)

I must say that running a local LLM with no tokens to buy changes your willingness to go slow and iterative with coding through an LLM, rather than alongside it.

Who knew writing a distributed db was hard?

@davidfowl Did the same. Takes very good prompt management and knowledge persistence.

I've been building a distributed systems with copilot playwright and aspire for a week without looking at any code to see if I can get it working well e2e... It works, but it was not easy. TL;DR building distributed systems is still hard AF 🙃

@davidfowl That said I do have issues with the js crashes and MCP inspector just doesn’t want me.

@davidfowl Currently having my agent use the mcp endpoint in aspire… while living in aspire. I may call it inception.net

Aspire skill 😍 #aspire #ai